[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"articles-1-4":3},{"code":4,"msg":5,"data":6},200,"成功",{"list":7,"total":48},[8,21,30,39],{"id":9,"title":10,"slug":11,"summary":12,"wordCount":13,"readingTime":14,"categoryId":15,"tags":16,"coverImage":16,"thumbnail":16,"status":17,"isTop":18,"isRecommended":18,"allowComments":18,"password":16,"viewCount":19,"likeCount":18,"commentCount":18,"publishTime":20},"2102319880609001474","工业物联网遥测双链路与按点位间隔入库技术实践","gongyewulianwangyaoceshuanglianluyuandianweijiangerukujishus","工业物联网遥测双链路与按点位间隔入库技术实践 文档类型：技术与架构实践分享 编写日期：2026-09-22 实践状态：SSE 实时推送与 InfluxDB 按点位间隔保存已完成现场功能验证 适用场景：设备高频遥测、实时大屏、时序数据持久化、Kafka 消费解耦 1. 摘要 工业物联网平台经常同时面对两类目标： 实时页面希望尽快看到设备最新值，不能被数据库写入速度影响； 时序数据库需要按照不同点位的业务价值控制保存频率，降低写入量和存储成本，同时不能静默丢失应保存的数据。 本实践最终采用“同一原始遥测 Topic、两个独立 Kafka 消费组”的双链路架构：实时链路只负责最新状态投影与 SSE 推送，持久化链路独立完成按点位间隔判断、批量写入 InfluxDB、水位提交和异常重试。两个链路共享无状态的标准化规则和点位映射缓存，但不共享消费位点、消费线程或下游成败状态。 现场功能验证表明：.....",12877,42,"0","",2,0,146,"2026-09-22 00:00:00",{"id":22,"title":23,"slug":24,"summary":25,"wordCount":26,"readingTime":27,"categoryId":15,"tags":16,"coverImage":16,"thumbnail":16,"status":17,"isTop":18,"isRecommended":18,"allowComments":18,"password":16,"viewCount":28,"likeCount":18,"commentCount":18,"publishTime":29},"2101860362313854978","语音转文字技术实践，分片、保活、声纹：内网转写系统的四个关键实现","yuyinzhuanwenzijishushijianfenpianbaohuoshengwenneiwangzhuanxiexitongdesigeguanjianshixian","经过半个月的时间不断学习和模型的训练，调试，整个语音识别这块的技术栈进行研究，本次主要是端到端，不调用互联网的API接口，自己通过小模型的学习，实现一套支持语音文件识别文字、识别声纹、杂波过滤、支持在线校对等于一体的语音识别文字，整个效果可以实现技术自主可控，成本低廉和可以正式使用的工具； 下面直接先看效果： ![](https:\u002F\u002Fseabrid.cn\u002Ffile\u002F2026\u002F2101862133727817728.png) ![](https:\u002F\u002Fseabrid.cn\u002Ffile\u002F2026\u002F2101862219169984512.png) ![](https:\u002F\u002Fseabrid.cn\u002Ffile\u002F2026\u002F2101862248702078976.png) ![](https:\u002F\u002Fseabrid.cn\u002Ffile\u002F2026\u002F2101863368358948864.png) > 只有 CPU、没有 ...",21155,70,186,"2026-09-21 00:00:00",{"id":31,"title":32,"slug":33,"summary":34,"wordCount":35,"readingTime":36,"categoryId":15,"tags":16,"coverImage":16,"thumbnail":16,"status":17,"isTop":18,"isRecommended":18,"allowComments":18,"password":16,"viewCount":37,"likeCount":18,"commentCount":18,"publishTime":38},"2069808550190837762","Qdrant 深度解析：面向向量搜索的高性能数据库","qdrant-shendujieximianxiangxiangliangsousuodegaoxingnengshujuku","## 一、什么是 Qdrant？ Qdrant 是一个**开源的向量数据库（Vector Database）**，专门用于存储、管理和检索高维向量数据。它通过高效的**近似最近邻搜索（ANN, Approximate Nearest Neighbor）**算法，实现毫秒级的相似度查询。 **核心定位：** > Qdrant = 向量存储 + 相似度搜索 + 过滤能力 **适用场景：** - 语义搜索（Semantic Search） - 推荐系统（Recommendation System） - 图像\u002F音频检索 - 大模型 RAG（知识检索增强） --- ## 二、为什么需要向量数据库？ 传统数据库（如 MySQL、PostgreSQL）主要处理结构化数据，而 AI 应用更多是**\"语义相似性\"**问题，例如： - \"找和这段话意思相近的内容\" - \"推荐和用户兴趣相似的商品\" **向量表...",2650,8,519,"2026-06-24 23:43:05",{"id":40,"title":41,"slug":42,"summary":43,"wordCount":44,"readingTime":45,"categoryId":15,"tags":16,"coverImage":16,"thumbnail":16,"status":17,"isTop":18,"isRecommended":18,"allowComments":18,"password":16,"viewCount":46,"likeCount":18,"commentCount":18,"publishTime":47},"2058861222097383425","Umi-OCR Docker 部署与 API 接口调用指南","umi-ocr-docker-bushuyu-api-jiekoudiaoyongzhinan","Umi-OCR 是一款免费、开源、可批量的离线 OCR 软件，支持截屏识别、批量导入图片、PDF 文档识别、排除水印\u002F页眉页脚、扫描\u002F生成二维码等功能。其 Docker 部署方案使得在 Linux 服务器或桌面环境下运行和集成变得非常便捷。 ### 一、Docker 部署 Umi-OCR #### 1. 部署前提 在开始部署前，请确保您的主机 CPU 支持 AVX 指令集，这是运行 Umi-OCR Linux 版本的必要条件。 ```bash lscpu | grep avx ``` 如果命令输出中包含 `avx` 和 `avx2`，则可以继续部署。 #### 2. 构建 Docker 镜像 首先，下载官方提供的 Dockerfile 并构建镜像。 ```bash # 1.",3556,11,619,"2026-05-25 18:42:19","43"]